AURORA

How far the bands separate — scales the angle assigned to each bin. 0 = mono; 100 = maximum unfurl.

The frequency→position curve of the fan. Centre = lows-centre / highs-edges; negative leans the width into the lows and reins the air back toward the centre. The character of the fan.

The spectral fan opens and closes by modulating the map. With SYNC it locks to tempo — the unfurl pulses with the track (capped under ~20 Hz, no audible AM).

AURORA's own net: ties the lows to the centre below a cut (50→400 Hz). The ILD is useless under ~1 kHz, so the bottom never widens itself apart.

The fan closes when the dry hits and reopens in the gaps — the width breathes with the performance instead of sitting static.

Dry/wet by power law (the dry is delay-aligned to the block). IN PHASE folds the wide image to mono-safe in one click.

Hann at 75% overlap with COLA — analysis/synthesis per block. No full phase vocoder (it never stretches time), which keeps the risk low.

Each bin gets an azimuth θ from the curve: its magnitude is split L/R by constant power and its phase is de-correlated per bin (opposite in L/R) — that's what opens the width, with no comb-filtering.

The mapping angle must vary slower than ~20 Hz or audible AM appears — Motion and the fast SYNC divisions are clamped under that bound. Honest by design.

Below the Mono Safe cut, bins collapse toward the centre — guaranteed mono sum, because real low-end width comes from de-correlation, not level.

An envelope-follower on the dry scales the global angle toward zero on transients — multiplicative on the angle, never touching phase or the power split.

Block latency (one STFT frame, N = 2048 samples) reported to the host and PDC-compensated — the trade-off for real spectral unfurling vs a zero-latency widener.